Obtaining highly accurate uniformity of response calibration measurements from the system as a whole may be very difficult, however, the high spatial frequency variation will mainly be restricted to the CCD chip, and the other components will generally contribute only low and medium frequency components. Hence, it is useful to calibrate the CCD chip uniformity of response separately (an easier experiment).
The effect of different temperatures and read-out speeds should not be too important. Nevertheless, the read-out conditions might as well be kept identical.
The CCD camera can be illuminated by an uniform source of visible light
e.g. an illuminated piece of paper. The exposure camera could be rotated
by 90
during the measurement several times to reduce any low
frequency unevenness in illumination, or such effects could be removed in
software. An appropriate dark count image should be subtracted from the
flat-field images, and several flat-field images should be averaged to
reduce statistical noise, although this should be less of a problem than
for the dark count images.
It may be useful to obtain reference images with and without X-rays. Probably reference images which include X-ray background are the best for subtracting from the fluorescent cell measurements. In this manner X-rays not coming from the source position should be eliminated from the flat-field response. Using a fluorescent cell only containing the solvent should allow structure coming from the solvent, e.g. the water ring, to be removed from the fluorescence flood-fields. Some scaling will be necessary to account for the absorption of the fluorescent sample. In the case of the diffraction data, a good reference image will be an empty capillary.
The detector dark count is assumed to vary from pixel to pixel and consist of a
constant and a time varying quantity. It is temperature dependent, so
the CCD camera should be at the temperature that subsequent data is going
to be taken. With the Photometrics camera this means -40C will
be used36.
The dark count characteristics change as a function of read-out speed, so this should be fixed. Again, with the Photometrics camera the read-out speed is fixed to 200kHz, whereas the Princeton Instruments system allows 50, 100, and 150 kHz.
The dark count images will be subject to statistical noise, which should be relatively appreciable, since (hopefully) the dark count images will not be very strong. To reduce this noise source, the average of a few different images should be used.
The simplest approach is just to integrate dark count images for exactly the same time period as the subsequent data images, and the flood-field images. However, if the build up of dark counts is really as simple as a constant plus a linear time dependent quantity, two different measurements at different time periods would be sufficient to obtain all these parameters for all different exposure lengths. This model could be tested by making dark current measurements for many different integration times, and seeing if the values do fit this simple model.
The grid should be carefully placed on the detector in a standard orientation (so that the orientation is reproducible). The sample to mask distance should be measured (as it is effectively the sample to detector distance for the corrected data).
The aim is to calibrate the the flat-field response of the detector to X-rays coming from the sample position. Owing to the lack of an isotropic X-ray source it is necessary to use a source with a anisotropic distribution and calibrate the distribution with a 2-theta scan of a 0-D detector.
In Steps 1d and 1f a collimator may
be used such that the 0-D scintillation counter (or similar detector) is
only sensitive to X-rays from the sample position. The counting time at each
point should be sufficient, so that counting statistics errors are well
below 1% . This will probably require a long scan time, so either the
beam decay needs to be monitored with an I
monitor, or the scan
needs to be carried out in both directions and averaged (more
sophisticated treatment would be possible, but averaging is probably
quite good enough given the ESRF beam decay37).
In Step 2e the flood-field illumination of the XRII will include illumination from sources other than the sample e.g. hutch background. We are not interested in the response from other sources, so we include Step 2f where the flood-field source is removed and only the X-rays from other sources are measured. By subtracting the two images (measured for equal times) the flood-field image we measure is that from the sample point alone (except for secondary scattering by the air in-between the sample point and the detector; this should be negligible).
In subtracting the reference image from the fluorescent cell image we are automatically taking into account the dark current. Again it may be useful to average several images to reduce counting statistics errors.
When correcting the diffraction patterns we will correct an image with both peak and background combined by the flat-field response to the sample position. Since subsequent software will separate the peak from the background, the fact that the background (which does not necessarily come from the sample position) is also corrected by the flat-field response to the sample position is unimportant.